Experimental and Statistical Assessment of Sand Concrete Incorporating Ceramic Waste Sand

2.1 Raw materials

The raw materials used in this study are shown in Figure 1. Portland composite cement (CPJ–CEM II 42.5 S-L), supplied by the Hadjar Essoud cement plant (Skikda, Algeria), was used as the main binder. This cement has an absolute density of 3.230 g/cm³ and complies with the requirements of standard EN 197-1 for CEM II type cements. The fine aggregates consisted of two types of sand: natural dune sand (DS) of a siliceous nature and 0/1 particle size class, sourced from the Oued Z'hor quarry, east of Skikda, and recycled sand from CWS of 0/4 particle size class, produced by grinding and screening ceramic sink scraps. A fine addition of limestone (L), collected from the filtration units of the Ben Azzouz quarry (east of Skikda), was also incorporated into the matrix. More than 80% of the limestone filler particles are less than 80 µm, which helps to improve the densification of the matrix. A high-performance water-reducing admixture, Master Glenium 26 (SP), was used to improve the workability of the mixtures without increasing the water/cement ratio. This is a polycarboxylate ether-based superplasticizer, supplied as a light brown liquid with a pH of 5.0 and a density of 1.07 g/cm³. Finally, tap water was used to mix all concrete formulations.

Figure 1. Raw materials used: (a) dune sand (DS); (b) ceramic waste sand (CWS), and (c) limestone filler (L)

Table 1 shows the main physical and chemical properties of the fine aggregates used. These data provide the basic characteristics needed to understand the influence of each component on the fresh and hardened properties of the sand concrete. The results presented in Table 1 reveal distinct physicochemical differences between the materials studied. DS has a mainly siliceous SiO₂ composition of 94.09% and a density of 2.630 g/cm³, while CWS is characterized by a significantly higher alumina Al₂O₃ content of 22.3% and a slightly lower density (around 2.435 g/cm³) than DS. The sand equivalent value of CWS and DS is 86% and 75%, respectively, indicating a higher level of cleanliness and lower clay content in the recycled ceramic material. The fineness modulus of CWS and DS are 3.12 and 1.6, respectively, thus suggesting that its incorporation may improve particle settling and potentially enhance the mechanical performance of sand concrete. In addition, CWS has a slightly lower water absorption of 2.80% compared to 3.21% for DS and a lower content of fines less than 80µm of 3.33% compared to 6.67% for DS, reflecting a less porous and more resistant aggregate. 

Table 1. Physical and chemical properties of sands

The results presented in Table 1 reveal distinct physicochemical characteristics among the investigated materials. DS exhibits a high siliceous composition SiO₂ of 94.09% and an absolute density of 2.630 g/cm³, whereas CWS is distinguished by a remarkable alumina content Al₂O₃ of 22.3% and a slightly lower absolute density of 2.435 g/cm³. The sand equivalent value of CWS and DS are 86% and 75%, respectively, indicating the higher cleanliness of the recycled ceramic material. The fineness modulus of CWS and DS is 3.12 and 1.6, respectively, suggesting that their incorporation can enhance packing density and improve mechanical strength. Furthermore, CWS exhibits slightly lower water absorption of 2.80% compared to 3.21% for DS, as well as a reduced fines content of 3.33% vs. 6.67% for DS, reflecting a less porous and more stable material. 

The limestone filler (L) confirms its carbonate nature with a CaCO₃ content of 84.60%, making it a suitable mineral additive for matrix densification. These differences indicate a complementary interaction between the materials, which can be exploited to optimize the formulation of durable, high-performance cementitious composites. Finally, the use of Master Glenium 26 superplasticizer (pH = 5.0, density = 1.070 g/cm³) allows for precise adjustment of the rheology of the mixture while maintaining mechanical strength and durability.

Figure 2. Particle size distribution of sands used

Figure 2 shows the particle size distribution curves for DS and CWS. DS is a siliceous material with a particle size of 0/1 mm and a fine grain size, with more than 80% of particles passing through a 0.5 mm sieve. This narrow distribution results in a low fineness modulus, limited compaction density, and higher water demand. Meanwhile, CWS has a granular class of 0/4 mm and a larger particle size distribution with a significant proportion of coarse particles (ranging from 1 mm to 4 mm). This better granular distribution promotes particle compaction, reduces voids, and strengthens the granular structure, thus contributing to the optimization of the internal structure and potentially improving the mechanical performance and durability of sand concrete [22].

2.2 Experimental program

The experimental program consisted of partially replacing DS with CWS in the composition of sand concrete, at volumetric substitution rates of 0%, 10%, 20%, 40%, and 60%. These replacement rates were chosen to cover a wide range of substitutions while remaining compatible with technical feasibility and current industrial practices. The main objective of this study is to evaluate the influence of CWS incorporation on the fresh properties (workability, bulk density, and entrained air content) and hardened properties (compressive strength, flexural strength, modulus of elasticity, and ultrasonic pulse velocity) of sand concrete. The reference mix (SC0) was designed according to the experimental procedure established in the Sablocrete project as reported in the work of Chanvillard and Basuyaux [23], whereas the other mixes (SCCW) were prepared by gradually replacing the natural DS with recycled ceramic sand (CWS) in the reference formulation. For all mixtures, several parameters were kept constant, including a water/cement ratio (W/C) set at 0.68. This value was chosen to ensure adequate workability of the sand concrete, which is characterized by a very fine granular structure and a large specific surface area. The selected W/C ratio complies with the sand concrete formulation principles proposed by Chanvillard and Basuyaux [23], which recommend relatively high-water contents for concrete incorporating DS. Maintaining a constant W/C ratio in all mixtures made it possible to evaluate in isolation the effect of replacing DS with ceramic CWS on the properties of fresh and hardened concrete. In addition, the use of a high water-reducing polycarboxylate-based superplasticizer (SP) at a dosage of 0.9% by weight of cement made it possible to obtain a satisfactory consistency while limiting the negative effects of the relatively high-water content on mechanical performance, as shown in previous studies on fine-grained concretes. The detailed proportions of the mixture are shown in Table 2.

In order to characterize the properties of sand concrete in its fresh and hardened states, a series of standardized tests was carried out. The fresh state tests included measuring the fresh density, carried out in accordance with standard NF EN 12350-6, to assess the compactness of the mixture and verify its conformity. Workability was evaluated using the Abrams cone slump test, in accordance with standard NF EN 12350-2, which indicates the consistency of the concrete and its ease of placement. In addition, the air content was determined using a pressure air meter, in accordance with standard NF EN 12350-7, which made it possible to quantify the volume of air in the mixture, a parameter known to influence both durability, particularly against freeze-thaw cycles, and mechanical performance. For the hardened state, prismatic specimens measuring 4 × 4 × 16 cm³ were cast in the laboratory at a temperature of 25 ℃ and relative humidity of 50% and hardened in water until the test was completed. The flexural strength was determined using a three-point flexural test in accordance with standard NF EN 196-1, after which the resulting half-specimens were subjected to a compression test, which allowed the flexural and compressive strengths to be evaluated simultaneously. The modulus of elasticity at 28 days was estimated using the empirical relationship proposed by Neville [24], which correlates the compressive strength and stiffness of the material. Finally, the internal quality and homogeneity of the concrete were evaluated using ultrasonic pulse velocity (UPV) measurements, performed on 15 × 15 × 15 cm³ cubic specimens after 28 days of curing in water, in accordance with standard NF EN 12504-4.

Table 2. Mix proportions of different concrete mixtures (kg/m³)

2.3 Statistical analysis and validation of response surface method models

2.3.1 Statistical modeling

In order to evaluate the substitution effect of DS by CWS on the performance of sand concrete, to identify the interactions between variables, and to optimize the mechanical and physical properties, statistical modeling was performed using RSM, based on a full factorial design. The experimental data were adjusted using least-squares regression to establish a polynomial model representing the material's behavior. This model's validity has been checked using analysis of variance (ANOVA) and prediction graphs. Finally, a comparison between the experimental results and the predicted values made it possible to evaluate the accuracy of the model and determine the optimal conditions.

2.3.2 Parameter estimation using the least squares method

The least squares method allows a mathematical model to be fitted with experimental data by minimizing the deviation between observed values and those predicted by the model [25]. Furthermore, the factorial design is an appropriate methodology for conducting experimental tests in a systematic manner, allowing the effects of different variables and their interactions to be evaluated. This approach is particularly useful in contexts where several factors may influence the measured response, and its ability to isolate individual effects is a major advantage in statistical research [26]. The experimental results can be expressed using the following mathematical Eq. (1):

$Y=\beta_0+\beta_1 X$     (1)​

where, Y, X, and β_0,1 are the response (e.g., strength, slump, etc.), the independent variables of studied factors (substitution rate in our case), and the model coefficients, estimated using the least squares method, respectively.

4.1 Linear regression modeling (RSM)

4.1.1 Fresh state

The statistical models at the fresh state obtained via RSM are given in Table 3. These models enable the effect of the substitution rate of DS by CWS on the fresh properties of sand concrete to be quantified. The estimated coefficients of the equations show a gradual decrease in density with increasing CWS rate, which is attributed to the lower density of the recycled material. The regression coefficient β₁ of -0.034 indicates a loss of approximately -0.034 g/cm³ for each 10% substitution, confirmed by a significant t-ratio of -4.46. In addition, the entrained air content shows a significant increase of +0.745%, due to the irregular and angular morphology of the CWS particles, which promotes air entrapment. The model predicts an increase of approximately +0.075% per 10% substitution increment, with a high t-ratio of 6.03, validating the robustness of the result. Workability shows a clear improvement with a higher CWS rate, due to better particle distribution and lower water absorption. The model indicates a gain of approximately +0.286 cm for each 10% substitution, with a t-ratio of 4.84. These effects are highly statistically significant (t-ratios > |2|) and highlight the potential of CWS to improve the fresh properties of sand concrete, while contributing to the sustainable recovery of ceramic waste.

Table 3. Statistical results of the sand concrete studied in the fresh state

Figure 10. Validation of RSM models in the fresh state: (a, d and g) correspond to the fitting curves of fresh density, air content, and workability respectively, as a function of the CWS substitution rate, (b, e and h) correspond to the correlations between experimental and predicted data for fresh density, air content, and workability, respectively, and (c, f and i) correspond to the random distribution of residuals for fresh density, air content, and workability, respectively

Figure 10 illustrates the validation of statistical models developed using RSM for the fresh properties of sand concrete incorporating CWS. Figures 10(a), (d), and (g) show the properties determined experimentally and those predicted for fresh density, air content, and workability, respectively, as a function of the CWS substitution rate. Figure 10(a) shows a clear linear decrease in fresh density with increasing CWS content, confirming the influence of the lower specific density of CWS compared to DS. This effect is well represented by the model, as shown by the close alignment between the experimental and predicted values in Figure 10(b) [47]. The corresponding residual plot (Cf. Figure 10(c)) shows a random distribution around zero within the ± 3σ confidence limits, demonstrating the homoscedasticity and reliability of the regression fit [48].

Furthermore, Figures 10(d) to (f) indicate a positive linear relationship between air content and substitution by CWS. The increase in entrained air is clearly illustrated in Figure 10(d). The excellent agreement between experimental and predicted values in Figure 10(e) and the absence of systematic bias in the residuals in Figure 10(f) further confirm the reliability of the model.

Figures 10(g), (h), and (i) present the results for workability. The strong positive correlation between slump values and CWS rate in Figure 10(g) indicates improved consistency with higher substitution rates. Figure 10(h) compares experimental and predicted values and shows points closely aligned along the midline, confirming the good predictive accuracy of the model. The residual analysis in Figure 10(i) shows a pattern of uniform dispersion within acceptable statistical limits, indicating that the normality assumption has not been violated and confirming the adequacy of the regression model.

In general, the findings 10 confirm the excellent predictive power and statistical validity of the RSM models for the fresh properties of sand concrete. The linearity, low residual dispersion, and narrow confidence intervals prove that the models effectively describe the experimental behavior and can be reliably used to predict fresh density, air content, and workability across the entire substitution range studied.

4.1.2 Hardened state

The hardened statistical models obtained via RSM are presented in Table 4. These models enable the effect of the substitution rate of DS by CWS sand on the mechanical and elastic properties of sand concrete to be quantified. The estimated coefficients of the equations reveal a constant improvement in strength and stiffness with increasing CWS content. Compressive strength shows a significant increase of +3.10 MPa for each 10% substitution, with a very high t-ratio of 7.36, confirming the high statistical significance of the effect. In addition, flexural strength shows a marked increase of +2.46 MPa for every 10% substitution, supported by a t-ratio of 10.28, indicating an even stronger influence of CWS on the tensile performance of the composite material. Also, the dynamic modulus of elasticity (Ed) shows a moderate increase of +0.935 GPa for every 10% substitution, with a positive regression coefficient and a t-ratio greater than the critical value of |2|, suggesting that adding CWS helps improve the stiffness and overall rigidity of the material. Conversely, the ultrasonic pulse velocity (UPV) shows a slight negative variation of –40.92 m/s related to each 10% substitution, accompanied by a low but significant t-ratio.

Table 4. Statistical results of the sand concrete studied in the hardened state

Figure 11 shows the validation of statistical models obtained using RSM for the properties of hardened sand concrete incorporating CWS. Figures 11(a), (b), and (c) illustrate the variation in compressive strength as a function of the CWS substitution rate. A clear positive linear trend is observed, indicating that the incorporation of CWS improves compressive strength. The close alignment between the predicted and observed data (Figure 11(b)) and the random distribution of residuals around the zero line within ±3σ (Figure 11(c)) confirms the accuracy and reliability of the regression model.

Figures 11(d), (e), and (f) show the results for flexural strength. The strong positive correlation (R² > 0.95) highlights the beneficial influence of CWS on the cementitious matrix. The predictive model is highly consistent, as the experimental values follow the predicted curve within the 95% confidence interval. Analysis of the residuals (see Figure 11(f)) confirms the absence of heteroscedasticity, thus validating the homogeneity of the data set.

Figures 11(g), (h), and (i) show a decrease in ultrasonic pulse velocity (UPV) with increasing CWS rate, suggesting a slightly more heterogeneous internal structure. However, UPV values remain within acceptable limits, reflecting good compactness and mechanical integrity of the material. The observed and predicted values (Figure 11(h)) show satisfactory correlation, and the residuals (Figure 11(i)) remain randomly dispersed without systematic deviation, confirming the relevance of the statistical model.

Figure 11. Validation of RSM models in the hardened state: (a, d, g, and j) correspond to the fitting curves of compressive strength, flexural strength, UPV, and Ed, respectively, as a function of the CWS substitution rate, (b, e, h, and k) correspond to the correlations between experimental and predicted data for compressive strength, flexural strength, UPV, and Ed, respectively, (c, f, i, and l) correspond to the random distribution of residuals for compressive strength, flexural strength, UPV, and Ed, respectively

Figures 11(j), (k), and (l) illustrate the model's performance for the dynamic modulus of elasticity (Ed). A slight but constant increase in Ed is observed with higher CWS incorporation. The regression curve corresponds well to the experimental data, and the residuals remain uniformly distributed, indicating that the model appears to be robust.

In general, the results presented in Figure 11 confirm the statistical robustness and goodness of fit of the RSM models applied to the hardened properties of sand concrete containing CWS. The linear relationships, the 95% narrow confidence bands, and the well-behaved residuals collectively indicate excellent predictive reliability and compliance with the fundamental assumptions of normality, independence, and homoscedasticity [47, 48]. These findings validate the models developed as effective tools for quantifying and optimizing the influence of CWS on the mechanical performance of durable sand concrete mixtures.

4.2 Significance of the models - analysis of variance

Given the importance of the CWS on the performance of the concrete in both its fresh and hardened states, a comprehensive statistical analysis was performed using analysis of variance (ANOVA). In the fresh state, the ANOVA results are presented in Table 5 and show that all the models developed are statistically significant, with p-values below 0.05. This confirms that the substitution rate of DS by CWS has a real and measurable influence on the various response variables. The coefficients of determination (R²), which ranged from 0.869 to 0.924, indicate that the models explain approximately 87 to 92% of the total variance in the experimental data, confirming an excellent level of agreement between the predicted values and those obtained experimentally. Furthermore, the adjusted R², which remains close to the unadjusted R², confirms the robustness of the regression models and prevents any overestimation due to unnecessary explanatory variables [49].

The predictive coefficients (Pred-R²), although slightly lower, remain within acceptable limits, confirming the satisfactory predictive capacity of the models. In addition, the root mean square error (RMSE) and the coefficient of variation (CV), which are less than 5%, provide good evidence of the high precision of the experimental data. Finally, the adequate precision values (Adeq Pr), which all exceed the recommended limit of 4, confirm a high signal-to-noise ratio in the models developed [50]. These statistical indicators collectively validate the reliability of the RSM approach for modeling and predicting the influence of CWS incorporation on the fresh properties of sand concrete [51].

Table 5. Analysis of variance (ANOVA) in the fresh states

Std. Dev: standard of deviation; C.V: coefficient of variation; PRESS: predicted residual error of sum of squares; Adj-R²: R² adjusted; Pred-R²: predicted R²; Adeq Pr: adequate precision.

The statistical evaluation of performance in the hardened state is presented in Table 6. The ANOVA results confirm that all models are statistically significant, with p values ≤ 0.05, indicating a significant effect of the CWS substitution rate on the responses studied. The coefficients of determination (R²) range from 0.772 to 0.972, demonstrating the models' strong ability to reproduce experimental variation. According to Schober et al. [52], R values greater than 0.9 indicate an excellent correlation between experimental and predicted results, which was the case for most models in this study.

The adjusted R² values remain close to the standard R² values, confirming the stability and robustness of the models while helping to avoid overfitting. The predictive coefficients (Pred-R²) show an excellent predictive accuracy for compressive strength (equal to 0.8237) and flexural strength (equal to 0.9234), highlighting their high potential for use in predictive applications and formulation optimization. Conversely, more moderate and weaker predictive capabilities are observed for UPV (Pred-R² = 0.4873) and Ed (Pred-R² = 0.0822), suggesting that these parameters may be influenced by additional microstructural factors that are not accounted for in the present model.

The RMSE values, which are generally low for the main mechanical responses, confirm the accuracy of the model adjustments. The coefficients of variation (CV), although variable depending on the property considered, remain within acceptable limits for mechanical testing on cementitious materials [53]. In addition, adequate precision values greater than 4 for all models, except for Ed, indicate a good signal-to-noise ratio and reinforce the reliability of the predictive equations developed.

Finally, these results demonstrate that the RSM approach is a robust and effective tool for modeling and optimizing the mechanical and physical performance of sand concretes incorporating CWS. It allows for a quantitative assessment of the influence of recycled fine aggregates on the structural behavior of cementitious materials, while contributing to sustainable construction practices [51]. Although there are some limitations in terms of dynamic properties, which may require additional modeling or experimental refinement, the overall predictive performance of the models developed confirms the potential of RSM as a decision-making method for the design of eco-efficient concrete formulations incorporating materials from the circular economy.

Table 6. Analysis of variance (ANOVA) in the hardened states

Std. Dev: standard of deviation; C.V: coefficient of variation; PRESS: predicted residual error of sum of squares; Adj-R²: R² adjusted; Pred-R²: predicted R²; Adeq Pr: adequate precision.

The authors would like to acknowledge the technical support received from the staff and facilities at the Laboratories of the Habitat and Construction (LHC-East) in Skikda, Algeria, and from the GICA Cement Company in Skikda, Algeria.